A Texture Based Pattern Recognition Approach to Distinguish Melanoma from Non-Melanoma Cells in Histopathological Tissue Microarray Sections

Elton Rexhepaj,Per-Henrik Edqvist,Michael Bergqvist,Fredrik Ponten,Margrét Agnarsdóttir,Emma Lundberg,Julia Bergman,Mathias Uhlén,William M Gallagher,Jürgen Schauber

doi:10.1371/journal.pone.0062070

Abstract

AimsImmunohistochemistry is a routine practice in clinical cancer diagnostics and also an established technology for tissue-based research regarding biomarker discovery efforts. Tedious manual assessment of immunohistochemically stained tissue needs to be fully automated to take full advantage of the potential for high throughput analyses enabled by tissue microarrays and digital pathology. Such automated tools also need to be reproducible for different experimental conditions and biomarker targets. In this study we present a novel supervised melanoma specific pattern recognition approach that is fully automated and quantitative.Methods and ResultsMelanoma samples were immunostained for the melanocyte specific target, Melan-A. Images representing immunostained melanoma tissue were then digitally processed to segment regions of interest, highlighting Melan-A positive and negative areas. Color deconvolution was applied to each region of interest to separate the channel containing the immunohistochemistry signal from the hematoxylin counterstaining channel. A support vector machine melanoma classification model was learned from a discovery melanoma patient cohort (n = 264) and subsequently validated on an independent cohort of melanoma patient tissue sample images (n = 157).ConclusionHere we propose a novel method that takes advantage of utilizing an immuhistochemical marker highlighting melanocytes to fully automate the learning of a general melanoma cell classification model. The presented method can be applied on any protein of interest and thus provides a tool for quantification of immunohistochemistry-based protein expression in melanoma.

Highlights

Antibody-based proteomics provides an advantageous strategy for biomarker discovery efforts [1]
A support vector machine melanoma classification model was learned from a discovery melanoma patient cohort (n = 264) and subsequently validated on an independent cohort of melanoma patient tissue sample images (n = 157)
Here we propose a novel method that takes advantage of utilizing an immuhistochemical marker highlighting melanocytes to fully automate the learning of a general melanoma cell classification model

Summary

Introduction

Antibody-based proteomics provides an advantageous strategy for biomarker discovery efforts [1]. Immunohistochemistry (IHC) is a well established and accepted assay for labeling a specific protein in tissue, provided the availability of validated antibodies towards the target of interest. IHC assays use an antibody to couple a candidate protein and a dye that makes the immunoreaction visible to the human eye in a microscope or a digital glass slide scanner. In routine IHC, hematoxylin a natural dye binding to nuclei acid, is added to the tissue specimen to highlight the cell nuclei in dark blue, cell cytoplasm and extra cellular matrix in light blue. TMAs have become an established and crucial component of the cancer biomarker discovery and validation pipeline

Methods

Results

Discussion

Conclusion